I am starting my Neural Network Project for the first time, I have a lot of expectations:
The main idea is primarily analyzing the pattern of my writings or drawings when using the mouse given a determined set of steps. I recorded myself writing the letter 'Z' and saving the direction of each stroke to a dataset of shape
(200, 50, 9). Example of an individual set:
[[0, 0, 0, 0, 0, 0, 0, 1, 0], ..., [0, 0, 0, 0, 0, 0, 0, 0, 1]],
(1, 9) vector is which direction the mouse is currently heading towards.
My network is fixed just like this:
_________________________________________________________________ Layer (type) Output Shape Param # ================================================================= lstm (LSTM) (None, 10, 50) 12000 _________________________________________________________________ dropout (Dropout) (None, 10, 50) 0 _________________________________________________________________ lstm_1 (LSTM) (None, 50) 20200 _________________________________________________________________ dropout_1 (Dropout) (None, 50) 0 _________________________________________________________________ dense (Dense) (None, 30) 1530 _________________________________________________________________ dropout_2 (Dropout) (None, 30) 0 _________________________________________________________________ dense_1 (Dense) (None, 9) 279 ================================================================= Total params: 34,009 Trainable params: 34,009
I divided each set into 10 sequences to make the LSTM work. and consecutively the output is the next (1, 9) vector to which the mouse should move.
However, I have hit a wall and the model does not seem to learn, I might suspect there is something wrong with my dataset. So I have come to clarify one thing:
- Is this dataset even trainable?, or in other words, I want to know whether there is an X model that is capable of predicting the next vector direction. So I just can keep trying and finding the optimal model or even keep cleaning the data. Or, I am missing a basic concept that makes the dataset lack information to be understandable to any model.
Thanks and regards.